Add Cosine Distance

[carterharrison]

I implemented Cosine Distance as a DistanceMetric. It is a common distance metric used for text embeddings. Let me know what I can do to get this merged!

[tom-whitehead]

Hey @carterharrison, thanks for contributing! My HDBSCAN implementation uses a KD Tree nearest neighbours algorithm to calculate distances to the kth (min_samples-th) neighbour. I've shied away from implementing cosine distance in the past as it
it's not commonly used with KD Trees. As I understand it, KD Tree NN is largely designed to work efficiently with distance metrics that satisfy the properties of Euclidean space and queries using cosine distance can become inefficient quickly, particularly as dimensionality increases. It is more common to use cosine distance with the Ball Tree NN algorithm (which I haven't been able to support yet).

That said, I tried this out on a couple of test datasets and the clusters looked good, albeit very similar results to Euclidean distance. As such I have nothing against merging it.

Have you tested it at all? If so, are you happy with the quality of clusters and performance?